In r266692, we made it possible to emit linkage-names only for inlined subroutines, putting the attribute
on the abstract origin DIE (which counts as the definition DIE). However, this fails when we emit a
declaration DIE prior to the definition DIE, because (a) when we process the declaration, we don't emit
the linkage name, because it's not an abstract origin; and (b) when we process the definition, we look at
the declaration metadata and see it has a linkage name, so we think it has already been emitted.
Look at the actual DIE, rather than the metadata, to see whether we've already emitted the linkage name.
I had thought about making this
and so not needing the extra static function, but actually looking into the DIE itself seemed more robust.